Temporary sharing prompts unrestrained disclosures that leave lasting negative impressions.

With the advent of social media, the impressions people make on others are based increasingly on their digital disclosures. However, digital disclosures can come back to haunt, making it challenging for people to manage the impressions they make. In field and online experiments in which participants take, share, and evaluate self-photographs ("selfies"), we show that, paradoxically, these challenges can be exacerbated by temporary-sharing media-technologies that prevent content from being stored permanently. Relative to permanent sharing, temporary sharing affects both whether and what people reveal. Specifically, temporary sharing increases compliance with the request to take a selfie (study 1) and induces greater disclosure risks (i.e., people exhibit greater disinhibition in their selfies, studies 1 and 2). This increased disclosure is driven by reduced privacy concerns (study 2). However, observers' impressions of sharers are insensitive to permanence (i.e., whether the selfie was shared temporarily versus permanently) and are instead driven by the disinhibition exhibited in the selfie (studies 4-7). As a result, induced by the promise of temporary sharing, sharers of uninhibited selfies come across as having worse judgment than those who share relatively discreet selfies (studies 1, 2, and 4-7)-an attributional pattern that is unanticipated by sharers (study 3), that persists days after the selfie has disappeared (study 5), is robust to personal experience with temporary sharing (studies 6A and 6B), and holds even among friends (studies 7A and 7B). Temporary sharing may bring back forgetting, but not without introducing new (self-presentational) challenges.

With the advent of social media, the impressions people make on others are based increasingly on their digital disclosures. Facebook alone has 1.86 billion active users who collectively post 136,000 photographs, 293,000 status updates, and 510,000 comments every minute (1), amounting to 4.75 billion pieces of content shared daily. Usage has been ever increasing, with 2016 representing a 15% increase from the year prior. People seem enamored with sharing photographs in particular; Instagram’s more than 400 million users post more than 80 million photographs each day (2).

Relative to traditional, offline forms of communication, there is an enhanced permanence to digital sharing. Disclosures are forever cataloged in the cloud, and, in the case of Tweets, also in the Library of Congress. Although individual offending posts can be deleted, it is all but impossible to expunge their every trace. Digital media content can be disseminated broadly—to friends, acquaintances, and strangers alike—with the click of a button. In turn, these audiences can effortlessly transmit this content to others. Moreover, it is often the most regrettable disclosures—compromising photographs posted in the heat of the moment—that are ripe for sharing and so perhaps are the hardest to undo.

The effective impossibility of undoing online disclosures presents new challenges to individuals in the digital age, especially when it comes to managing the impressions they make (3). Disclosures can come back to haunt, even those that seemingly disappear seconds after first exposure. In one case, flight attendants were fired for posting derogatory comments about their employer (4). In fact, a recent poll indicated that 93% of hiring managers check candidates’ social media activity, with the discovered information often hindering the chance of getting hired (5).

Given these dangers, one might wonder why people share in the first place. One reason is that self-disclosure confers benefits. Confiding in others is associated with better health (6) and professional benefits. Also, perhaps because it is a means of achieving connection with others—a fundamental human motivation (7)—self-disclosure confers psychological benefits, such as intimacy (8, 9) and liking (10, 11). Moreover, neuroscientific research suggests self-disclosure is intrinsically rewarding (12). However, people also have a desire for privacy (13, 14) and for good reason: Privacy is integral to human development (15). Because these desires—the desire for privacy and the desire to disclose—often run in opposition, honoring both simultaneously is a challenge, especially given the permanence of digital disclosures.

Enter temporary sharing: New technologies that place expiration dates on disclosures may help resolve the tension (16, 17). Snapchat, the photograph-sharing application (hereafter, “app”) wherein photographs and messages disappear after the recipient has viewed them has more than 150 million daily users (18). Facebook and Instagram also have popular temporary-sharing features; for example, Instagram Stories, on which posts automatically vanish after 24 h, has more than 250 million daily users (19, 20). From a narrow perspective, these technologies would seem to be a panacea, honoring both the desire to divulge and the desire for privacy. After all, content that no longer exists cannot come back to haunt.

Or can it? A broader, behavioral scientific perspective suggests that temporary sharing may not be the cure-all that it may seem to be. Impressions are sticky (21, 22), even when incorrect. Thus, an indiscreet, temporarily shared photograph may make an impression that persists beyond its short life. Moreover, observers may attribute the indiscretion to the (bad) judgment of the sharer rather than being warranted by the temporariness of the sharing platform. Indeed social psychologists have documented an actor–observer asymmetry in attribution: Observers tend to attribute behavior to actors’ enduring personality characteristics as opposed to situational influences; actors tend to do the opposite (23–28). Therefore, people’s impressions of sharers may be driven by the content of the photograph and not by sharers’ choice of sharing medium. This attribution pattern would pose a challenge in self-presentation via temporary sharing media, especially if the attribution by observers is unanticipated by sharers. Potentially compounding this issue, ephemerality, in its capacity to assuage privacy concerns, may increase disclosure, and of sensitive information in particular (13, 29). Consistent with this idea, computer-mediated settings, which feel private, in part because they reduce feelings of public self-consciousness, increase self-disclosure and decrease socially desirable responses (30–33).

We test this account in an experimental paradigm in which participants were asked to take and share a “selfie” (a photograph of themselves, taken by themselves), expecting others to view it. The present research therefore also contributes to the emergent science of the psychology of photo-taking (34, 35). We focus on the sharing of visual content (selfies) both because of its pervasiveness and because impressions are strongly affected by appearance. However, our predictions and their theoretical underpinnings also apply to other types of disclosures (e.g., text-based disclosures such as status updates, video posts, and so forth).

First, we predict that, relative to permanent sharing, temporary sharing affects both whether and what people reveal. Specifically, we predict that temporary sharing increases compliance with the request to take a selfie (hypothesis 1a, study 1) and causes people to take greater disclosure risks, which we operationalize by the disinhibition they exhibit in their selfies (hypothesis 1b, studies 1 and 2). Second, we predict that this increased disclosure is driven by reduced privacy concerns (hypothesis 2, study 2). Third, we predict observers’ impressions of sharers are based on the uninhibitedness of the selfie and are insensitive to permanence (i.e., whether the selfie was shared temporarily or permanently) (hypothesis 3, study 4). As a result, sharers of uninhibited selfies, induced by the promise of temporary sharing, will come across as having worse judgment than those who share relatively discreet selfies (hypothesis 4, studies 1, 2, and 4–7). Finally, we predict that this attributional pattern will be (i) unanticipated by sharers (hypothesis 5a, study 3); (ii) persistent after the selfie has disappeared (hypothesis 5b, study 5); (iii) robust to personal experience with temporary sharing (hypothesis 5c, studies 6A and 6B); and (iv) manifest even among friends (hypothesis 5d, studies 7A and 7B).

Results and Discussion

Study 1 was a field study that tested whether temporariness impacts disclosure (hypotheses 1a and b) and the impressions made upon others. Two research assistants approached people (n = 296; mean age = 25 y, SD = 5.4 y; 55% female) on a university campus, alternating in the use of a temporary-sharing vs. control solicitation script. The people were asked to take and share a selfie using a “Moment Machine” (36, 37)—a networked display application that allows passersby to take a photograph that is displayed on monitors across campus. Participants were made aware that their selfies would be posted on the monitors and on a publically accessible Facebook page. Those in the temporary condition were further informed, truthfully, that selfies would be posted temporarily, for 1 h. After data collection, research assistants coded the selfies for disinhibition and rated the sharers’ judgment. The procedural details, manipulations, measures, comprehension checks, pretests, and data exclusions for all studies are outlined in .

Temporary sharing made people 1.2 times more likely to take and share a selfie relative to those in the control condition [compliance: 70.00% in temporary vs. 57.55% in control, χ2(1) = 4.85, P < 0.05]. As for photograph content, an intent-to-treat analysis including participants who did not take a selfie indicated that temporariness made people 3.44 times more likely to take uninhibited selfies relative to the control condition [meantemporary = 52.00% vs. meancontrol = 15.11%, χ2(1) = 43.60, P < 0.005, Cohen’s d = 0.84, 95% CI (0.60, 1.08)]. There were various subtypes of uninhibited photographs: 71% contained a hand gesture; 28% had an unusual or silly face; 13% used a prop; 5% depicted unrestrained actions.

Temporary sharers were perceived as having worse judgment (meantemporary = 3.00, SD = 1.02) than those in the control condition [meancontrol = 3.88, SD = 0.98, F(1, 183) = 34.86, P < 0.005, Cohen’s d = −0.87, 95% CI (−0.57, −1.18)]. Selfies from the temporary condition were also rated more likely to go viral [meantemporary = 2.79, SD = 0.90 vs. meancontrol = 2.21, SD = 0.78; F(1, 183) = 20.71, P < 0.005]. To the extent that it is possible to “hack” temporary sharing (e.g., by taking a screenshot on Snapchat), this pattern implies an irony: Content that makes a person come across as having bad judgment may be particularly prone to being shared by others.

Study 2 was an online experiment in which we tested a mechanism hypothesized to underlie the disinhibition prompted by temporary sharing: dampened privacy concerns (hypothesis 2). Study 2 also featured several procedural enhancements relative to study 1: first, before taking a selfie, sharers were made explicitly aware that others from the same population would rate them based on their selfie. Second, we designed the experiment to induce equal selfie-sharing compliance across conditions, enabling us to document observers’ impressions of sharers with greater internal validity. As a result, we necessarily focused on the nature of the selfie shared (i.e., testing hypothesis 1b, whereas study 1 was optimized to test hypothesis 1a). Third, we obtained quality-of-judgment ratings from a larger sample of observers.

Participants [n = 428 US users of the Amazon Mechanical Turk website (hereafter “MTurkers”); mean age = 33 y, SD = 10.2 y; 42% female] took and shared a selfie via their webcams. Before they did so, we told them that other MTurkers would rate their photographs, randomizing whether we told participants that the photograph would be available temporarily or permanently. Next, participants reported their privacy concerns and then took and shared a selfie. Compliance was high and equal across conditions: 74.64% in the permanent-sharing condition, 76.26% in the temporary-sharing condition; χ2(1) = 0.15, not significant (NS). The selfies were coded for disinhibition as in study 1. A separate group of MTurkers (n = 71) rated the sharers’ judgment.

An intent-to-treat analysis indicated that participants were 1.52 times more likely to depict disinhibition in the temporary-sharing condition than in the permanent condition [meantemporary = 45.21% vs. meanpermanent = 29.67%, χ2(1) = 11.01, P < 0.005, Cohen’s d = 0.32, 95% CI (0.13, 0.52)]. Privacy concerns mediated the relationship between temporary sharing and selfie disinhibition: Temporariness decreased privacy concerns (βtemporary = −0.70, SE = 0.22, P < 0.005), which in turn accounted for increased disinhibition (βprivacy = −0.05, SE = 0.01, P < 0.005). The bootstrapped indirect effect is significant: β = 0.04, 95% CI (0.01, 0.08), SE = 0.02. includes a conceptual replication of these findings (study S1). Content coding indicated that the uninhibited photographs contained an unusual or silly face (56%), a hand gesture (39%), prop use (24%), nudity (6%), and drug use (3%).

Temporary sharers were viewed as having worse judgment than permanent sharers [meantemporary = 3.59, SD = 1.20; meanpermanent = 3.93, SD = 1.12, t(321) = 2.68, P < 0.01, Cohen’s d = −0.30, 95% CI (−0.08, −0.52)]. This effect held when controlling for rater fixed effects (βtemporary = −0.35, cluster robust SE = 0.12, P < 0.01) and standardizing ratings within raters (βtemporary = −0.27, cluster robust SE = 0.08, P < 0.005).

Studies 1 and 2 show that temporary sharing dampens privacy concerns (hypothesis 2), increasing people’s willingness to take and share selfies (hypothesis 1a) and to portray themselves as uninhibited in those selfies (hypothesis 1b). However, sharers of uninhibited selfies, induced by the promise of temporary sharing, came across as having worse judgment than those who shared relatively discreet selfies (hypothesis 4). Study 3 tested whether sharers anticipate these consequences (hypothesis 5a).

In study 3, participants (n = 200 US MTurkers, mean age = 37 y, SD = 12.5 y, 39% female) were randomized to imagine they were either about to share a selfie (sharers) or to receive a selfie from someone (observers). Participants were told that the selfie was risqué; it had a bit of nudity in it (see study S2 for a conceptual replication invoking a different, “very silly,” selfie). Next, we described two possible platforms on which the selfie could be shared, differing in permanence (i.e., Snapchat vs. iMessage). Both sharers and observers were asked whether the sharer would make a better impression if they sent the selfie via Snapchat, via iMessage, or if the platform would not matter. In this and all subsequent studies, multiple-choice tests administered after the outcome measure confirmed that participants had noticed the condition-specific information ().

Whereas most sharers (53.5%) believed sending the selfie via Snapchat would make the better impression, only 31.4% of observers agreed [χ(1) = 10.19, P < 0.005]. In contrast to sharers’ common intuition, observers most commonly indicated the platform would not matter (Fig. 1 and Fig. S1). Thus, study 3 suggests an actor-vs.-observer asymmetry: Sharers believe temporary sharing improves the impression they make on others (hypothesis 5a); observers believe their impression of sharers is unaffected by the platform.

Fig. 1.

Sharers commonly think they will make a better impression if they send an uninhibited selfie temporarily; observers think the platform temporariness does not matter (study 3). Bars sum to 1 within color.

Fig. S1.

Sharers have the common intuition that sending a selfie on a temporary platform (i.e., Snapchat) will make the best impression, whereas observers most commonly say that the platform would not impact their impression of the sharer (study S2). Bars sum to 1 within color.

Study 4 assessed the accuracy of these intuitions, testing whether observers temper their judgments of sharers when sharing is temporary (hypothesis 3). Although we obtained observers’ impressions of the sharers in studies 1 and 2, unlike these studies, in study 4, observers were made aware of the platform—temporary vs. permanent—on which the selfie had been shared. Also, in study 4 we necessarily manipulated selfie disinhibition independently from temporariness.

Participants (n = 339, mean age = 26 y, SD = 5.1 y; 52% female) were students at the same university as in study 1 and were shown a selfie that another student had taken on the Moment Machine. We randomized whether participants (i) viewed an uninhibited selfie or a control selfie and (ii) were told that the sharer had posted it temporarily or permanently. The latter manipulation was pretested to ensure participants noticed this information. Participants rated the sharer’s judgment.

Participants deemed sharers who appeared uninhibited as having worse judgment than those not appearing uninhibited [meanuninhibited = 3.68, SD = 1.52; meancontrol = 4.50, SD = 1.49, F(1, 337) = 24.59, P < 0.005, Cohen’s d = −0.54, 95% CI (−0.32, −0.76)]. These judgments did not depend on the platform [F(1, 335) = 0.05, NS], and the effect held within the temporary-sharing condition [meanuninhibited = 3.71, SD = 1.38; meancontrol = 4.48, SD = 1.66), t(145) = 3.01, P < 0.005] (Fig. 2).

Fig. 2.

Impressions of sharers are based on the content of the selfie and are insensitive to permanence (study 4). Error bars indicate SEMs.

Study 4 shows that observers’ impressions of sharers are based on the uninhibitedness of the selfie and are insensitive to permanence (hypothesis 3), pointing to the inaccuracy of sharers’ belief that others will view their disinhibition as appropriate for the ephemeral medium. A conceptual replication reported in the invoking a different platform (Instagram) showed the same result (study S3, Fig. S2). Study 5 tested whether a temporarily shared, uninhibited selfie makes an impression that persists beyond its brief life span (hypothesis 5b).

Fig. S2.

Perceived quality of judgment was based solely on whether the sharer appeared uninhibited in their selfie on Instagram (study S3). Error bars represent SEMs.

Study 5 was a longitudinal experiment spanning 3 d. On day one, participants (n = 403 US MTurkers; mean age = 36 y, SD = 11.6 y; 55% female) were randomized to view either an uninhibited or a control selfie, were told it had been shared temporarily via Instagram Stories, and rated the sharer’s judgment. On the next 2 d, participants reevaluated the sharer, this time from memory (the selfie was not redisplayed). Participants were randomized to reevaluate on either day two or day three. Thus, each participant rated the sharer twice, once alongside the selfie on day one and again from memory either 1 or 2 d after exposure. Most participants (82.88%) completed the second evaluation (NS by condition).

Upon viewing the selfie, participants deemed sharers of uninhibited selfies as having worse judgment than those of control selfies [time 1: meanuninhibited = 3.52, SD = 1.45; meancontrol = 4.91, SD = 1.29), t(401) = 10.21, P < 0.005; d = −1.02 95% CI (−0.81, −1.23)]. This effect held when participants later simply recalled their impression of the sharer, whether they did so 1 d later [time 2, day 2: meanuninhibited = 4.20, SD = 1.45; meancontrol = 4.84, SD = 1.13), t(167) = 3.23, P < 0.005] or 2 d later [time 2, day 3; meanuninhibited = 4.04, SD = 1.30; meancontrol = 4.85, SD = 1.15, t(159) = 4.21, P < 0.005]. Convergent results from more sophisticated regression analyses are given in and Table S1. The difference between impressions of sharers of uninhibited vs. not-uninhibited photographs attenuated between exposure and reevaluation, but the magnitude of this attenuation was equivalent whether reevaluation was 1 or 2 d after the initial exposure (Fig. 3). A conceptual replication of this study is given in (study S4, Fig. S3).

Table S1.

The less-than-stellar impression that an uninhibited, although temporarily shared, selfie makes on others lasts beyond its short life span (study 5)

	Model 1		Model 2		Model 3
DV: Quality of judgment	Coefficient	SE	Coefficient	SE	Coefficient	SE
Re- (vs. immediate) evaluation	0.176*	0.07	−0.095	0.089	−0.065	0.086
Uninhibited (vs. control)	−1.026***	0.127	−1.202***	0.206	−1.168***	0.215
Reevaluation after 2 d (vs. after 1 d)	−0.061	0.125	0.062	0.178	0.095	0.190
Uninhibited × reevaluation			0.596***	0.141	0.528**	0.174
Reevaluation × 2 d			−0.018	0.140	−0.083	0.176
Uninhibited × 2 d			−0.244	0.253	−0.313	0.298
Uninhibited × reevaluation × 2 d					0.139	0.283
Constant	4.839***	0.110	4.920***	0.127	4.905***	0.128
N	664		664		664
R2	0.138		0.152		0.152

DV, dependent variable; *P < 0.05, **P < 0.01, ***P < 0.005.

Fig. 3.

The less-than-stellar impression that an uninhibited, although temporarily shared, selfie makes on others lasts beyond its short life span (study 5). Error bars indicate SEMs.

Fig. S3.

The less-than-stellar impression that an uninhibited, although temporarily shared, selfie makes on others persists after it has disappeared (study S4). Error bars represent SEMs.

Study 5 suggests that the impression made by a temporarily shared uninhibited selfie persists beyond its short life (hypothesis 5b). Studies 6A and 6B tested whether observers’ assessments of sharers’ quality of judgment are tempered by personal experience with temporary sharing (hypothesis 5c).

In study 6A, we first described the temporary sharing features of Snapchat, Instagram, and Facebook and then asked participants (n = 400 US MTurkers, mean age = 36 y, SD = 11.5 y; 47.25% female) whether they had used such a feature (64.00% answered “yes”). Next, participants were randomized to view either an uninhibited selfie or a control selfie, were told that it had been shared temporarily, and rated the sharer’s judgment. Replicating studies 4 and 5, participants deemed sharers who appeared uninhibited as having worse judgment than those not appearing uninhibited [meanuninhibited = 3.74, SD = 1.56; meancontrol = 5.01, SD = 1.34, F(1, 398) = 77.23, P < 0.005; Cohen’s d = −0.88, 95% CI (−0.67, −1.09)]. These ratings were not tempered by experience with temporary sharing [F(1, 396) = 0.45, NS]; the effect held among those with such experience [meanuninhibited = 3.85, SD = 1.65; meancontrol = 5.05, SD = 1.37; t(254) = 6.35, P < 0.005].

Whereas study 6A measured experience with temporary sharing, study 6B (n = 396 US MTurkers, mean age = 33 y, SD = 9.2 y; 34% female) manipulated it: half of the participants were randomized to first upload and temporarily share a selfie. Most (82.90%) complied, although results are intent-to-treat. Next, participants were randomized to view either an uninhibited or a control selfie and were told that the sharer had posted it temporarily. Participants deemed sharers who appeared uninhibited as having worse judgment than those not appearing uninhibited [meanuninhibited = 4.00, SD = 1.72; meancontrol = 5.35, SD = 1.29, F(1, 394) = 80.13, P < 0.005; Cohen’s d = −0.90, 95% CI (−0.69, −1.11)]. Replicating study 6A, these ratings were not tempered by experience with temporary sharing [F(1, 392) = 0.80, NS]; the effect held among those with such experience [meanuninhibited = 4.08, SD = 1.78; meancontrol = 5.30, SD = 1.30, t(191) = 5.49, P < 0.005].

Studies 6A and 6B suggest personal experience with temporary sharing does not lead observers to factor the situational influence—the ephemerality of the medium—into their impressions (hypothesis 5c). So far, our tests of hypotheses 4 and 5 featured observers evaluating potentially unknown fellow students (study 4) or strangers (studies 5 and 6), a common occurrence. Attesting to this point, we conducted surveys in which we asked people about their social media networks (). Results indicated that users frequently encounter posts from strangers (Fig. 4). Looking within each of the three top platforms, many have set their profiles to public (Snapchat: 29%; Facebook: 22%; Instagram: 58%). As for the composition of users’ networks within each platform, a nontrivial proportion is comprised of strangers (Snapchat: 17%; Facebook: 38%; Instagram: 52%) in addition to acquaintances (Snapchat: 29%; Facebook: 40%; Instagram: 22%) and friends (Snapchat: 53%; Facebook: 22%; Instagram: 27%) (Fig. S4).

Fig. 4.

Self-reported frequency of encountering strangers’ posts.

Fig. S4.

Percentage of close friends, acquaintances, or strangers among friends on platform.

Nonetheless, studies 7A and 7B answer two outstanding questions: whether people view sharers of uninhibited photographs as having bad judgment even when those sharers are friends and, if so, whether this effect holds when sharing takes place on temporary media (hypothesis 5d). These studies test the impact of two independent variables, sharer type (friend vs. stranger) and selfie type (degree of disinhibition), on perceived sharer judgment.

Study 7A employed a naturalistic setting in which Instagram users (n = 400 US MTurkers, mean age = 29 y, SD = 7.7 y; 61% female) rated someone who had temporarily shared a selfie on Instagram Stories. On Instagram, users can access most others’ posts, enabling us to randomize participants to view a temporarily shared selfie of a stranger or a friend. The other independent variable was measured: Participants indicated the degree of disinhibition in the selfie. For the outcome measure, participants rated the sharer’s judgment. Administration order was counterbalanced and did not matter. According to observers, the more uninhibited the sharer, the worse the sharer’s perceived judgment (βuninhibited = −0.23, SE = 0.07, P < 0.005, Table S2, Model 2). This result did not depend on whether the sharer was a friend or stranger (βuninhibited × friend = 0.02, SE = 0.07, NS), although, perhaps not surprisingly, friends’ judgment was rated higher than strangers’ judgment (βfriend = 0.21, SE = 0.07, P < 0.005). It also held after controlling for demographics ( and Table S2, Model 3).

Table S2.

The negative relationship between perceived quality of judgment and the degree of disinhibition of the selfie was robust to whether the selfie came from a stranger or friend (study 7A)

DV: Quality of judgment	Model 1		Model 2		Model 3
	Coefficient	SE	Coefficient	SE	Coefficient	SE
Degree of disinhibition	−0.236***	0.070	−0.228**	0.070	−0.214**	0.070
Friend (vs. stranger)			0.205**	0.070	0.204**	0.070
Degree of disinhibition × friend			0.021	0.070	0.005	0.070
Age					0.025**	0.009
Gender					−0.051	0.143
Education					−0.107	0.099
Constant	5.148***	0.070	5.148***	0.069	4.770***	0.401
N	400		400		400
R2	0.028		0.049		0.068

DV, dependent variable; Degree of disinhibition and friend are standardized; **P < 0.01, ***P < 0.005.

Whereas study 7A measured disinhibition, study 7B manipulated it. Participants (n = 499 US MTurkers, mean age = 34 y, SD = 9.8 y; 41% female) imagined they encountered a selfie temporarily shared by either that friend or a stranger that was either uninhibited or not uninhibited. Participants rated the sharer’s judgment. Participants deemed sharers whom they supposed to have posted an uninhibited selfie as having worse judgment than those they supposed to have posted less uninhibited selfies [meanuninhibited = 4.33, SD = 1.24; meancontrol = 5.04, SD = 1.15, F(1, 497) = 43.75, P < 0.005; Cohen’s d = −0.59, 95% CI (0.42, 0.77)]. Replicating study 7A, this pattern did not depend on friendship status [F(1, 495) = 0.62, NS], although friends were deemed to have better judgment than strangers [meanfriend = 4.90, SD = 1.24; meanstranger = 4.48, SD = 1.21, F(1, 497) = 15.35, P < 0.005]. Thus, although the psychology of first impressions differs from that of existing relationships (38), studies 7A and 7B suggest that, at least in the domain of temporary sharing, both friends and strangers are penalized for uninhibited behavior.

Conclusion

People increasingly share personal information over social media platforms on which posts are only temporarily available. Such technologies would seem to be a panacea, simultaneously honoring two often-conflicting desires: the desire to disclose and the desire to protect one’s privacy. The present research points to a different conclusion: Temporary sharing can exacerbate the challenge of self-presentation in the digital age.

We documented one psychological driver behind the capacity for temporary sharing to induce disclosure: dampened privacy concerns. Future research could explore additional, complementary mechanisms underlying this phenomenon. For example, the tendency to honor sunk costs (39, 40) could make a person who has decided to use a temporary-sharing medium feel compelled to “make good” on that choice by sharing content they would not dare to share permanently. Future research might also test whether our findings extend to situations in which people post information about others: If a person surreptitiously takes and shares (if only temporarily) a photograph of someone acting uninhibitedly, who suffers the judgment penalty—the sharer, or the unwitting subject of the photograph? (The latter, we suspect). We also explored one facet of self-presentation: perceived quality of judgment. In addition to exploring how temporary sharing affects the different dimensions on which a person is perceived (e.g., warmth, competence), future work might also explore how temporary, as opposed to permanent, platforms affect the sharer’s enjoyment of that experience.

Given the documented pitfalls of temporary sharing, future research might explore other interventions designed to help people better manage the digital impressions they make—for example, by preventing people from making uninhibited disclosures in the first place. Cooling-off periods could be helpful in this respect: Users could opt-in to a feature that inserts a delay between the time when they press “post” and the time when that post appears (during that interval they could change their mind). For example, upon detecting inebriation, the “Drunk Text Savior” app disables texting, requiring would-be sharers to answer math questions correctly to restore functionality.

Relatedly, although social media are increasingly becoming broadcast tools, sharers do not always treat them as such, sometimes acting as if they are sending private letters when really the situation is more akin to sending postcards on which messages are in plain sight. By design, social media would seem to create an illusion of intimacy. It is as if, at the moment of divulgence, sharers have a specific target in mind—a person or group to whom they envision themselves to be disclosing—to the neglect of the truly broad scope of the audience. Embarrassment, or worse, can ensue. In one case, a Facebook user made a post complaining about her boss, forgetting that he was a Facebook friend; she was subsequently fired (41). Just-in-time interventions, delivered the moment before posting, could prompt users to consider audience scope (although they may introduce new issues, such as excessive self-censoring).

In sum, the use of temporary-sharing technologies does not, in and of itself, make a person come across as having bad judgment. Instead, it is the tendency for temporary sharing to induce risky disclosure (via assuaged privacy concerns), combined with the fact that observers’ impressions of sharers are based on the way those sharers look in the photographs and are insensitive to sharing platform choice, that produces this pattern. Temporary sharing may bring back forgetting, but not without introducing new (self-presentational) challenges.

Materials and Methods

Informed consent was obtained from all participants, and the Institutional Review Boards of the Università della Svizzera Ialiana and Harvard University reviewed and approved the materials and procedures. Data collection-stopping rules were preset (). Stimuli and nonidentifiable alphanumeric data are available at https://osf.io/qaw2v/.

Study 1.

Procedure.

After sharing the selfie, participants completed a survey and were entered into an iPad raffle. The survey included demographics (as did all subsequent studies), a request for permission to use the selfie in follow-up studies (we excluded those who declined or took a photograph that was not a selfie, n = 7), and two additional measures ().

Selfie coding.

Selfies including at least one of the following were coded as uninhibited: a silly or unusual face (e.g., sticking out the tongue); a hand gesture (e.g., a stranglehold); using an object as a prop (e.g., “attacking” the camera with an umbrella); unrestrained action shots (e.g., jumping); nudity; or drug use (smoking, drinking). Interrater agreement on the presence of disinhibition was 91% (κ = 0.80, z = 13.71, P < 0.005); disagreements were resolved by a third coder. Two other coders rated each sharer (“I think the person who shared the photograph has good judgment”) on a scale of 1 (strongly disagree) to 5 (strongly agree) (r = 0.40, P < 0.005). We took the average of raters’ assessments of each sharer. Two other coders rated whether each selfie “is likely to go viral” on a scale of 1 to 5. Coders in this and all studies were blinded to study conditions.

Study 2.

We sought to induce uniformly high compliance by stating during recruitment that participants would need a webcam. Participants could earn a bonus payment of up to $1.00 based on how others rated their photograph ().

Temporariness manipulation.

Participants received a description of either temporary or permanent sharing () that noted either that the raters could view their photograph “only once, for a maximum of 10 seconds,” or “for as long as they want.”

Privacy concern.

Participants were asked to respond to the statement “In thinking about what kind of photograph to upload, I am concerned about my privacy” on a scale of 1 (strongly disagree) to 7 (strongly agree).

For selfie coding, we used the same procedure as in study 1 (agreement = 95%, κ = 0.90, z = 16.11, P < 0.005).

Judgment ratings.

Judgment ratings were as in study 1, but a 1–7 scale was used (and also was used in studies 4–6). Each rater was randomly assigned a set of selfies to rate ().

Study 3.

Outcome measure.

After reading introductory text describing their role (), actors [observers] were asked: “Do you think you [the person] would make a better impression on the person you’re sending your risqué selfie to [in sending you the risqué selfie] if you [they] sent the selfie via Snapchat? Via iMessage? Or would it not matter?”

Check.

Most (88%, NS by condition) identified that sharing is permanent on iMessage and temporary on Snapchat.

Study 4.

Pretest.

Participants (n = 100, same population as in study 4) viewed the introductory text from either the temporary or permanent condition, read filler information, and then were quizzed on whether the text had invoked temporary or permanent sharing (pass rate: 90%, NS by condition).

Stimuli.

We used a pool of 20 selfies, generated by 10 students (same population as the participants), each of whom took two Moment Machine selfies: a control (e.g., smiling) photograph and an uninhibited one (silly/unusual face: n = 5; gesture: n = 7; prop: n = 3; action: n = 2; these sum to greater than 10 because some photographs had multiple markers of disinhibition).

After reading a description of the Moment Machine (), participants were told: “The person in the photograph below shared it temporarily [permanently].”

Study 5.

Stimuli (also for study 6A and 6B).

Participants were randomized to view one of 10 uninhibited (silly/unusual face: n = 6; gesture: n = 7; prop: n = 2; action: n = 1; some had multiple markers of disinhibition) or 10 not-uninhibited selfies from previous participants who had agreed to let us use them.

Most (93%, NS by condition) identified that the selfie had been shared “only temporarily (i.e., via Instagram Stories)” as opposed to “permanently (i.e., via Instagram).”

Study 6A.

Experience measure.

After reading a description of temporary sharing (), participants were asked: “Have you ever used such a feature? i.e., have you ever shared content temporarily via social media?” The response scale was binary, yes/no.

Most (94%, NS by condition) correctly identified whether the selfie had been shared “only temporarily (i.e., via Instagram Stories)” vs. “permanently (i.e., via Instagram).”

Study 6B.

Experience manipulation.

Participants in the experience condition were told “This photo will not be stored permanently on our servers. Just like when sharing via Snapchat, it will only be temporarily available for a few moments and self-destructs afterwards.”

Most (91%, NS by condition) correctly identified whether the selfie they evaluated had been shared via “ephemeral social media (e.g. Snapchat)” or “regular text messaging application (iMessage).”

Study 7A.

Selfie lookup.

We prompted participants to type a letter (randomized by participant) into the Instagram search screen, generating a list of friends’ and strangers’ accounts with that letter. Participants were instructed to visit the Instagram Stories page of either the first friend or stranger on this list (randomized between subjects), to click through the Story, stopping at the first selfie they saw; if there was no selfie, they were to select the second friend [stranger] and continue until they found one.

Selfie disinhibition measure.

Participants responded to the item: “the person in the selfie appears…,” on a response scale of 1 (not at all uninhibited) to 7 (extremely uninhibited). We validated this measure in a pretest: As in study 7A, US MTurkers (n = 87) found a selfie on Instagram Stories and rated its disinhibition. Participants also provided information for us to download the selfies. Two coders rated the selfies on the same disinhibition scale; their ratings correlated with participants’ (r1 = 0.37, P < 0.005; r2 = 0.34, P < 0.005).

Because of the noise introduced by the realism of this context and because people already have a lot of information about their friends, the judgment measure focused on the posting decision: “Based on the decision to post this selfie, I think the person in the selfie has good judgment” on a response scale of 1 (strongly disagree) to 7 (strongly agree). Study 7B uses a more conservative measure.

Checks.

We asked participants whether they and the person they had evaluated followed each other (mutual following is presumably less common for strangers). In the friend condition, 63% indicated mutual following, vs. 37% in the stranger condition [χ(1) = 34.34, P < 0.005]. Most correctly identified whether they had (i) rated a friend vs. a stranger (93%, NS by condition) and (ii) looked up a selfie on Stories as opposed to the permanent feed (88%, NS by condition).

Study 7B.

Sharer type manipulation.

Half of participants first entered the name of a friend, which was then piped into the instructions: “Imagine you are on Instagram and see a selfie posted by your friend, name [a stranger]. Further imagine your friend, name [the stranger] shared the selfie only temporarily, by using Instagram Stories. This means that the selfie can only be viewed temporarily; the photo self-destructs after 24 h.”

Selfie type manipulation.

Selfies were described textually only, by having participants suppose the sharer either looked “normal—i.e., smiling” or “very uninhibited,” by “making a very silly face.”

Judgment measure.

Participants responded to the statement “If I saw that my friend, name [a stranger] had shared a normal-looking [very silly] selfie on Instagram Stories, I would think that they have good judgment” on a scale of 1 (strongly disagree) to 7 (strongly agree).

Most correctly identified whether the sharer (i) was “a stranger” or “your friend” (95%, NS) and (ii) “was simply smiling” or “looked very silly” (78%, NS).

Additional Details of Studies in the Main Text

Sample Sizes.

We prespecified when data collection would end (i.e., the decision to stop collecting data was independent from results; we did not analyze the data until after data collection for the given study had been completed).

Moment Machine studies.

For study 1, we collected as much data as we could in the 7 d during which we had been allocated use of the Moment Machine. For study 4, we emailed prospective participants (students at the university at which the Moment Machine selfies were taken) and collected as much data as we could in the 7 d after we sent the email (at which point students left for holidays).

Online studies.

As a general rule of thumb, following recent thinking on sample size (datacolada.org/18), we sought to obtain a minimum of 100 participants per cell. As a result, in studies 3, 5, 6A, 6B, and 7A we prespecified the sample size to 100 per cell. For studies 2 and 7B, we prespecified even larger sample sizes. In study 2, this was because we were concerned about having sufficient power to detect mediation of a binary outcome measure, so we increased the sample size, prespecifying 400 participants. Although we only needed 100 participants per cell for study 7B, this study was administered with (i.e., in the same MTurk Hit as) another, unrelated study for cost-saving purposes. This unrelated study required a sample size of 500. Slight deviations from the target numbers and the actual number are caused by idiosyncratic differences in how survey “completes” are registered in MTurk vs. the survey software we used to collect the data. We report the results using all observations. As a robustness check, however, when the obtained number was larger than the target number, we reran the analyses, restricting the data to the first x responses, where x represents the prespecified target number. The results hold when we do this check.

Raffle details.

We informed participants how they could increase their chances of winning the iPad as follows. First, they would receive one raffle ticket for completing the survey. Second, they received an extra ticket for taking and sharing a photograph of themselves. Third, they received an extra ticket for each comment or “like” that their selfie received on Facebook within the first hour of its posting.

Additional measures.

The survey concluded with two additional, exploratory measures: how much regret participants had over the selfie they had chosen to share and their privacy concerns (both NS by condition). In study 2 we used a stronger set-up to assess the possible mediating role of privacy concerns. Specifically, in this experiment the mediator was measured in the proper causal sequence (as opposed to being measured at the end, as in study 1), and there was greater power to detect mediation because more participants shared a selfie. We then conducted a conceptual replication of study 2 (study S1) in which we again found evidence of this mechanism. In addition to the raters that assessed the sharers’ judgment, two additional raters assessed the extent to which each selfie “is likely to go viral,” i.e., whether they thought others might be inclined to disseminate the photograph further, using a scale of 1 to 5.

Bonus payment details.

Participants received a bonus payment based on how other MTurkers rated their photograph, on a scale of 1 to 5. Specifically, participants were (truthfully) told that 25% of them would be randomly selected to receive a bonus based on this rating ($0.20 for a 1 rating, $0.40 for a 2 rating, $0.60 for a 3 rating, $0.80 for a 4 rating, and $1.00 for a 5 rating).

Details on quality-of-judgment evaluation.

In total, we obtained 1,124 ratings from 71 raters. Each rater rated an average of 15.83 photographs (SD = 3.92; range, 2–20). Photographs were randomly allocated to raters with those photographs with fewer ratings having a higher chance of being picked (weighted random selection). Each rater could evaluate the same photograph only once. Photographs were evaluated by 3.48 raters (SD = 0.70; range, 2–6) on average.

Introductory text.

The introductory text described temporary and permanent sharing. Participants in the temporary condition were told:

“Photo-taking task: Your Photo will Only Temporarily be Stored

On the next page we will ask you to take a photo of yourself. When you take your photo, please be aware that this photo will not be stored permanently on our servers. It will only be temporarily available and deleted afterwards. The Turkers rating your photo may view it only once, for a maximum of 10 s. They cannot download the photo. Nobody will be able to access the photo once this study is completed. Please consider this information when taking and sharing your photo in this study.”

Participants in the permanent condition were told:

“Photo-taking task: Your Photo will be Stored Permanently

On the next page we will ask you to take a photo of yourself. When you take your photo, please be aware that this photo will be stored permanently on our servers. That is, the photo will be stored for at least 10 y on our servers. The Turkers rating your photo may view it for as long as they want and as many times as they want. They may also download the photo and review it even after the study is completed, even after 10 y from now. Please consider this information when taking and sharing your photo in this study.”

Study 3: Introductory Text.

The introductory text described temporary sharing and was tailored to the participant’s role (sharer vs. observer). Sharers were told:

“Please imagine that you are about to share a photo of yourself (i.e., a selfie) via a mobile application. The selfie that you’ve chosen to share is risqué… you are showing a bit of nudity in it. On Snapchat, all shared photos are available temporarily. So if you chose to share your selfie on this application, it would automatically disappear once the recipient has viewed it. On iMessage, all shared photos are available permanently. So if you chose to share your selfie on this application, it would not disappear once the recipient has viewed it.”

Similarly, observers were told:

“Please imagine that someone has decided to send you a photo of themselves (i.e., a selfie) via a mobile application. The selfie that the person chose to share is risqué… the person is showing a bit of nudity in it. On Snapchat, all shared photos are available temporarily. So if the person chose to share their selfie on this application, it would automatically disappear once you had viewed it. On iMessage, all shared photos are available permanently. So if the person chose to share their selfie on this application, it would not disappear once you had viewed it.”

Study 4: Introductory Text.

The introductory text described temporary and permanent sharing. Participants in the temporary condition were told:

“A Selfie that was Shared Temporary on the Moment Machine

Now we would like to show you a photo that has been taken using the Moment Machine. The photo that you are about to see was shared only temporarily on the Moment Machine. Before sharing, we told the person that the photo will be deleted from the Moment Machine after one hour. This means that the selfie could only be viewed temporarily; the photo was deleted by us after 1 h. After these instructions, the person took and shared the photo.”

Participants in the permanent condition were told:

“A Selfie that was Permanently Shared on the Moment Machine

Now we would like to show you a photo that has been taken and shared using the Moment Machine. Note that photos are available permanently on the screens after sharing.”

The introductory text described temporary sharing on Instagram Stories. All participants were told:

“Imagine Someone Shared a Selfie (e.g., via Instagram Stories)

Imagine that you are on Instagram and see a selfie posted by a stranger. Further imagine that the stranger shared the selfie only temporarily, by using Instagram Stories. This means that the selfie can only be viewed temporarily; the photo self-destructs after 24 h.”

Regression analysis.

We ran a regression among participants who provided both the initial and the reevaluation (n = 334) with quality of judgment as the dependent variable and the between-subject manipulations uninhibited (vs. control) and evaluation gap (reevaluation after 2 d vs. 1 d) as independent variables. We also controlled for the reevaluation (vs. immediate evaluation) of the selfie within-subjects (i.e., we have two observations for quality of judgment per individual, the immediate evaluation and the later reevaluation). Upon viewing the selfie (i.e., immediate evaluation), participants deemed sharers of uninhibited selfies as having worse judgment than sharers of control selfies (βuninhibited = −1.03, SE = 0.13, P < 0.005) (Table S1, model 1). This effect significantly attenuated when participants simply recalled their impression of the target, without being shown the photographs again (βuninhibited × reevaluation = 0.60, SE = 0.14, P < 0.005) (Table S1, model 2). The magnitude of the attenuation does not differ between participants who reevaluated 1 d vs. 2 d after exposure (βuninhibited × reevaluation × 2 d = 0.14, SE = 0.28, P > 0.05) (Table S1, model 3). Thus, although the uninhibited photographs were no longer available to observers, they made enduring impressions.

Study 6A: Description of Temporary Sharing.

First, we described temporary sharing:

“Many social media platforms now enable temporary sharing. For example:

On Snapchat, users can share content temporarily. When this feature is enabled, the content (e.g., photograph) disappears immediately after the recipient has viewed it.

On Instagram, users can use a feature called “Instagram Stories” that shared photographs only temporarily. When using Instagram Stories, the shared content can only be viewed for 24 h; after that time, it disappears.

On Facebook, users can set a temporary profile picture. When using this feature, the temporary profile picture is only visible for the period of time specified by the user; after that point the profile photo automatically reverts back to a different profile picture of the user’s choosing.”

Second, we told participants that the selfie they were about to evaluate was shared temporarily:

“Imagine Someone Shared a Selfie Temporarily on Instagram Stories

Instagram is a social media platform that enables people to share photos. Instagram Stories is a feature on this platform that enables users to share photos only temporarily – when using Instagram Stories, the shared content can only be viewed for 24 h; after that time, the content disappears.”

We prescreened individuals who had their cell phone with them and who owned an Instagram account. During the survey, we asked them to take out their cell phone and open the Instagram app. In a two-cell experiment, we manipulated whether they had to evaluate a selfie of a friend or a stranger. We told them to go to the search screen and type in a random single letter of the alphabet. This prompts Instagram to load a list of Instagram account names containing that letter, showing a mix of people they already follow and complete strangers. In the friend condition, we told them to pick the first friend that appeared in this list. In the stranger condition, we told them to pick the first stranger on the list. Then, we asked them to open that person’s Story (by clicking on/tapping the profile photograph), search for the first selfie of the person in the story, look at it for a moment, and indicate how they perceive the person’s degree of disinhibition and quality of judgment. If the person had no active Story or no selfie in it, we told them to find the next person (stranger or friend depending on the experimental group) in the list. If they could not find a selfie from a person on the current list, they were asked to type in the next letter of the alphabet and repeat the above steps. To increase the chances that participants took the task seriously, we asked them to indicate the username of the person on the selfie and the time when the selfie was posted on Instagram.

We applied regression analysis for hypothesis testing with quality of judgment as the dependent variable and the degree of disinhibition and the friend-vs.-stranger manipulation as independent variables (both standardized). As expected, we find the coefficient for degree of disinhibition to be negative and significant (βuninhibited = −0.24, SE = 0.07, P < 0.005) (Table S2, model 1). This effect is not significantly moderated by whether the selfie comes from a friend or a stranger (βuninhibited × friend = 0.02, SE = 0.07, P > 0.05) (Table S2, model 2) and is robust to adding demographic controls (Table S2, model 3). Intuitively, selfies of friends are evaluated more positively (βfriend = 0.20, SE = 0.07, P < 0.01) (Table S2, models 2 and 3).

Study 7B: Full Text of Manipulation.

Participants randomized to the friend condition were first asked to “enter the name of one of your friends” into an accompanying text box. This text entry—the friend’s name—was then piped into all subsequent references to that friend on the next page, which is where the rest of the manipulation-specific text appeared, and the outcome measure was administered. Specifically, both manipulations (of sharer type: stranger vs. friend and selfie type: uninhibited vs. not uninhibited) were reflected in the following text that was printed immediately above the outcome measure. The text below reflects the stranger, uninhibited condition. In the not-uninhibited condition, the italicized text was replaced with the text shown in brackets. In the friend condition, all instances of the term “a stranger” and “the stranger” were replaced with “your friend, name” where “name” was the actual name the given participant had entered on the previous page.

“Imagine a stranger shared a selfie via Instagram Stories.

Imagine that you are on Instagram and see a selfie posted by a stranger. Further imagine that the stranger shared the selfie only temporarily, by using Instagram Stories. This means that the selfie can only be viewed temporarily; the photo self-destructs after 24 h.

What do you think of that person’s quality of judgment?

Further, imagine a stranger has posted a selfie in which they look very uninhibited [normal] – i.e., the stranger is making a very silly face [smiling].”

Conceptual Replications

Study S1: Conceptual Replication of Study 2.

Study S1 is a conceptual replication of study 2 with a different operationalization of the dependent variable (sharing). It provides converging evidence of the process underlying temporary sharing’s capacity to induce disclosure. Participants (n = 200 US MTurkers, mean age = 36.03 y, SD = 10.70 y, 48.00% female) indicated how uninhibited a selfie they would choose to share as a function of the temporariness of the sharing medium. Specifically, participants imagined that they were about to share a selfie online using social media. Between-subjects, we randomized the permanence of the sharing medium on which they imagined sending it: Snapchat (temporary condition) vs. iMessage (permanent condition). To confirm that participants had processed the condition-relevant information (i.e., whether the platform was a temporary- vs. permanent-sharing medium), at the end of the study we asked participants a multiple-choice question: “This survey was about a specific mobile application. A core feature of this application was that: A (any photo shared would be available to whom I share it only temporarily), B (any photo shared would be available to whom I share it permanently), C (none of the above).” Overall, 92% of participants passed this check (NS by condition).

Next, participants were asked: “Given that you will be sharing the photo on Snapchat [iMessage], what type of selfie would you choose to share? Specifically, how uninhibited a selfie would you be inclined to share?,” using a five-point scale: 1 (not uninhibited), 2 (slightly uninhibited), 3 (moderately uninhibited), 4 (very uninhibited), and 5 (extremely uninhibited). Before asking this question, we told participants: “By ‘uninhibited’ we mean things like: making a goofy face, making a profane gesture, showing nudity, and so forth.” Next, we measured privacy concerns. Participants indicated their agreement with the statement: “I would be concerned with my privacy when sharing a [level of disinhibition participant previously indicated] photo on Snapchat [iMessage]” using a seven-point scale with endpoints labeled 1 (strongly disagree) and 7 (strongly agree). Finally, participants predicted recipients’ attributions of their disclosure behavior on a seven-point scale by completing the sentence: “I think that recipients of my photo will think I chose to share a [level of disinhibition participant previously indicated] photo because…” on a slider scale with endpoints labeled “…of my personal desire to send the photo” and “…the app—Snapchat [iMessage] —warranted it.”

Participants indicated that they would be more likely to share uninhibited selfies in the temporary condition (meantemporary = 2.50, SD = 1.20; i.e., slightly to moderately uninhibited) relative to the permanent condition [meanpermanent = 2.09, SD = 1.15; i.e., slightly uninhibited, t(198) = 2.49, P < 0.05]. As in study 2, regression results indicate that temporariness is a significant predictor of privacy concerns (βtemporary = −0.61, SE = 0.27, P < 0.05) and that privacy concern is a significant predictor of the degree of disinhibition (βprivacy = −0.20, SE = 0.04, P < 0.005). These results support the mediational hypothesis. Approximately 27% of the proportion of the total effect is mediated by the privacy concern measure. We tested the indirect effect using a bootstrap estimation approach with 10,000 samples and found that it was significant [β = 0.11, 95% CI (0.02, 0.24), SE = 0.06]. Finally, participants were more likely to think that recipients would attribute their disclosure choices to the platform warranting it in the temporary condition (meantemporary = 2.97, SD = 1.75) than in the permanent condition [meanpermanent = 2.48, SD = 1.48, t(198) = 2.14, P < 0.05].

Study S1 provides converging evidence of the basic finding that temporary sharing induces greater disclosure risks (hypothesis 1) using methods complementary to studies 1 and 2. Specifically, study S1 provides support for hypothesis 1 using a different, more ecologically valid operationalization of the independent variable (i.e., invoking Snapchat vs. iMessage platforms) as well as a different operationalization of the outcome measure (sharers state for themselves the degree of disinhibition they would exhibit, as opposed to uploading selfies and having those selfies coded by research assistants, as in studies 1 and 2). Study S1 also provides converging evidence for a psychological mechanism underlying the capacity for temporary sharing to induce disclosure: reduced privacy concerns (hypothesis 2). Finally, study S1 suggests that sharers may believe others will attribute their disinhibition to situational factors (i.e., the temporariness of the medium) as opposed to their disposition (hypothesis 5a).

Study S2: Conceptual Replication of Study 3 with Alternate Selfie Type (Silly Instead of Risqué).

In study S2, participants (n = 400 US MTurkers, mean age = 34.91 y, SD = 11.35 y, 52.75% female) were randomized to imagine that they were either about to share a selfie (sharers) or to view a selfie that someone had sent them (observers). Participants were further told that the selfie was uninhibited; they looked very silly in it. Next, we described two possible platforms on which the selfie could be shared, differing in permanence (i.e., Snapchat vs. iMessage), as in study 3. Both sharers and observers were then asked for their intuition about whether the sharer would make a better impression if they sent the selfie (i) via Snapchat (ii) or via iMessage, or (iii) whether the platform did not matter, as in study 3.

Whereas most sharers (53.30%) believed that sending the selfie via Snapchat would make the better impression, only 37.32% of observers agreed [χ(2) = 10.47, P < 0.005]. In contrast to the most common intuition of the sharers, observers most commonly indicated that the application on which the selfie was sent would not affect their impression of the sharer (Fig. S1).

Study S2 replicates the actor-vs.-observer asymmetry with a different type of selfie shared: sharers believe that temporary sharing will improve the impression they make on others (hypothesis 5a), whereas observers believe that their impression of sharers will be unaffected by the sharing platform.

Study S3: Conceptual Replication of Study 4 with Instagram.

Study S3 is a conceptual replication of study 4 within Instagram, invoking Instagram Stories in the temporary condition and the regular Instagram feed in the permanent condition. Participants (n = 407 US MTurkers, mean age = 35.79 y, SD = 18.78 y, 49.63% female) were randomized to either a temporary-sharing or permanent-sharing condition and to view either an uninhibited selfie or a control selfie.

Selfie manipulation.

Participants were randomized to view either an uninhibited selfie or a control selfie. As in study 4, to reduce the chance that observed effects are driven by idiosyncratic features of specific selfies, each participant was randomly presented with one selfie out of a pool of 20 possible selfies, 10 of which had been coded as uninhibited and 10 as not-uninhibited. The selfies were of individuals in a previous study who had agreed to let us use them. Thus, in the uninhibited selfie condition, each participant was randomly presented with one of the 10 uninhibited selfies; similarly, in the control selfie condition, each participant was randomly presented with one of 10 control selfies.

Participants in the temporary condition were asked to “Imagine that you are on Instagram and see a selfie posted by a stranger. Further imagine that the stranger shared the selfie only temporarily, by using Instagram Stories. This means that the selfie can only be viewed temporarily; the photo self-destructs after 24 h.” In the permanent condition, participants were asked to “Imagine that you are on Instagram and see a selfie posted by a stranger. Further imagine that the stranger shared the selfie permanently, by using regular Instagram. This means that the selfie can be viewed permanently.”

Pretest of temporariness manipulation.

As in study 4, participants (n = 96 US MTurkers, the same population as study S3) viewed the introductory text from either the temporary or permanent condition, read filler information, and then were quizzed on whether the text had invoked temporary or permanent sharing (pass rate: 87%, NS by condition).

Results.

Participants deemed targets who appeared uninhibited in their selfies as having significantly worse judgment than those not appearing uninhibited [meanuninhibited = 3.49, SD = 1.56; meancontrol = 4.84, SD = 1.28, F(1, 405) = 89.59, P < 0.005]. These judgments were not tempered by sharing mode; targets appearing uninhibited were deemed as having just as bad judgment regardless of whether sharing was temporary or permanent [F(1, 403) = 1.43, P > 0.05] (Fig. S2).

Study S4: Conceptual Replication of Study 5 with Snapchat.

Study S4 was a two-part study. First, participants (n = 213, mean age = 35.80 y, SD = 11.82 y, 44.60% female) were presented with either an uninhibited selfie or a control (i.e., not-uninhibited) selfie from a stranger and were told that it had been shared over Snapchat and that they would rate the person in the selfie. As in study 5, each participant was randomly presented with one selfie out of a pool of 20 possible selfies, 10 of which had been coded as uninhibited and 10 as not-uninhibited. The selfies were of individuals in a previous study who had agreed to let us use them. Thus, in the uninhibited selfie condition, each participant was randomly presented with one of the 10 uninhibited selfies; similarly, in the control selfie condition, each participant was randomly presented with one of the 10 not-uninhibited selfies.

Specifically, participants were told: “In a few moments, we will ask you to evaluate the quality of judgment of another person, based on a photo that the person shared online. When stating your evaluation of the photo in a few moments, please imagine that the person in the photo sent it to you via ephemeral social media (e.g., Snapchat). Ephemeral means that after having received the photo, you can only view it temporarily and it self-destructs after a few moments.” Next, participants rated the target’s judgment using the same item as in study 5.

Two days later, participants were asked to complete a follow-up survey in which we told them: “In part 1 of this study, we showed you a photo of a person and then asked you to evaluate this person’s quality of judgment” and asked them to “indicate the person’s quality of judgment.” Participants were further told: “We will not show you the photo again, simply respond based on your memory of the person you saw.” Participants then rated the target’s quality of judgment using the same item as in part one. The response rate was high: 83.1% of those who had completed part one also completed part two [NS between conditions, χ(1) = 0.27, P > 0.05].

Upon viewing the selfie (i.e., part one), participants deemed targets of uninhibited selfies as having worse judgment than those of control selfies [T1: meanuninhibited = 3.74, SD = 1.54; meancontrol = 5.39, SD = 1.20, t(211) = 8.64, P < 0.005], an effect that held 2 d later, when participants simply recalled their impression of the target, without being shown the photos again [T2: meanuninhibited = 4.13, SD = 1.65; meancontrol = 5.18, SD = 1.18, t(175) = 4.82, P < 0.005] (Fig. S3). Thus, although the uninhibited photos were no longer available to observers, the impressions that they made persisted beyond their short life.

Follow-Up Surveys

We ran four surveys to explore whether it is common to see content from strangers on social media.

Follow-Up Survey 1: Procedures.

The results of the first survey are summarized in Fig. 4 in the main text. Here are the procedures for this survey. We collected information from 100 US MTurkers (mean age = 32.87 y, SD = 9.24 y, 61% female). First, participants indicated which of the top three social media platforms they use: Snapchat, Facebook, or Instagram. Next, for each platform a given participant indicated using, we asked the following two questions:

“When you open/visit Snapchat [Facebook, Instagram], how frequently do you see posts made by strangers (people you’ve never met in person and are not actually friends with in real life)?” Response options were:

Never (0% of the time I visit Snapchat [Facebook, Instagram])

Rarely (20% of the time I visit Snapchat [Facebook, Instagram])

Sometimes (40% of the time I visit Snapchat [Facebook, Instagram])

Often (60% of the time I visit Snapchat [Facebook, Instagram])

Very often (80% of the time I visit Snapchat [Facebook, Instagram])

Always (100% of the time I visit Snapchat [Facebook, Instagram])

We further asked whether their profile is set to public or private. This item was customized to reflect the possible settings of the given platform, presented in the order in which they were listed on the platform. To create a binary measure of whether a given profile was private vs. public, we recoded the responses. Response options shown in boldface type were coded as indicative of a public profile:

Snapchat: “Please look at your profile privacy settings for your Snapchat account. Is your profile set to “everyone” (anyone can see what you post), “my friends” (only your accepted friends can see what you post), or “custom”? My Snapchat privacy setting is:

Everyone

My friends

Custom

I don’t know”

Facebook: “Please look at your profile privacy settings for your Facebook account. Is your profile set to “Public” (anyone can see what you post), “Friends” (only your accepted friends can see what you post), or another custom setting? My Facebook privacy setting is:

Public

Friends

Friends except acquaintances

Only me

I don’t know”

Instagram: “Please look at your profile privacy settings for your Instagram account. Is your profile set to public (anyone can see what you post) or private (only certain people can see what you post)? My Instagram privacy setting is:

Public

Private

I don’t know”

Participants selecting the “I don’t know option” were provided detailed instructions, including screenshots, of where to find this information on the given platform.

The survey concluded with basic demographic questions.

Follow-Up Surveys 2–4: Procedures.

To learn what proportion of online friends (i.e., connections) consist of strangers, we conducted three additional surveys, one each for Snapchat (n = 108 US MTurkers, mean age = 30.39 y, SD = 7.52 y, 59% female), Facebook (n = 102 US MTurkers, mean age = 34.52 y, SD = 10.69 y, 51% female), and Instagram (n = 100 MTurkers, mean age = 31.95 y, SD = 8.53 y, 51% female). We asked users of these platforms to categorize their connections as stranger, acquaintance, or friend. Doing so is not as simple as asking people to click on their friend list and categorize each friend, because

Users typically have many connections (in the hundreds or even thousands), so it was not feasible to ask participants to categorize the population of connections. Therefore, for each platform, we asked each user to categorize only a sample of their connections; however,

We could not simply ask people to categorize, say, the first 10 friends in their list, because the default sample in which friends are listed is nonrandom (typically closer contacts appear first). Therefore, for each platform we devised a procedure for participants to generate an effectively random sample of connections; the details were necessarily slightly different by platform, but the overall approach was the same: Based on the assumption that friend closeness is independent from names (i.e., people do not base their decisions on whether to be friends with a person based on the letter that person’s name starts with), we asked participants to categorize a sample of their friends generated by alphabet letter. All instructions were accompanied by annotated screen shots depicting the instruction in action on the given platform. Here is the procedure, by platform:

On Snapchat, we guided participants to find their friend list sorted in alphabetical order and asked participants to categorize the first 10 contacts on this alphabetized list. On Facebook and Instagram, it is not possible to sort the contact list alphabetically, so we needed to direct participants to do this through search: On Facebook, we guided participants on how to find their friend list and to enter a single letter into the search box for the friend list (randomized between-subjects to be one of three, arbitrarily chosen letters: r, b, or m). Doing so causes Facebook to generate the list of all the given user’s connections whose first and/or last names begin with the given letter. We asked participants to categorize all contacts generated by this search procedure. On Instagram, we guided participants on how to find the list of people they are following and to enter randomly generated letters into the search box. Because on Instagram, as on Twitter, connections do not need to be mutual, in the sense that following someone is independent from being followed by that person, we also asked participants to do the same categorization task a second time, this time for the people following them. In our data analysis we focus on the former, because it was measured first and it is more relevant to our goal of ascertaining whether people frequently encounter content by strangers (thus on Instagram focusing on the people a user follows rather than on the people who are following the given user).

In each survey, once the list had been generated, participants were told: “We want to know what proportion of people’s Snapchat [Facebook, Instagram] friends are… -family or close friends, -distant friends or acquaintances, -people you’ve never met.” Specifically, we asked them to categorize each person on the list they generated, indicating how many people on that list:

They have never met

They consider to be either distant friends or mere acquaintances

Are either family members or close friends

Each survey ended with basic demographic questions.

Summary of Results of Follow-Up Surveys Not Already Reported in the Main Text.

Results indicate that strangers make up a nontrivial proportion of participants’ networks (Fig. S4). Moreover, 29.20% of the Snapchat users indicated they have set their profile to public, as had 21.6% on Facebook and 57.80% on Instagram.